Project Summary/ Abstract Liver cancer is one of the leading causes of cancer deaths with rising incidence in the U.S and worldwide. Yttrium-90 microspheres radioembolization, or Selective Internal Radiation Therapy (SIRT), is a treatment in which a catheter inserted in the patient's hepatic artery delivers radioactive 90Y microspheres to the liver. It is increasingly utilized to treat patients with unresectable liver tumors in second or third line, but some of its potential to improve overall survival is still untapped. The major obstacle in making SIRT more efficient is the treatment planning. It consists in selecting the 90Y activity to inject based on the estimated dose to the tumor and organs-at-risk. The problem is that the dose calculation is highly unreliable and does not include important parameters, such as well-known non-uniformities or the injection point. As a result, physicians often choose very conservative dosage to limit toxicity at the expense of the tumor(s) dose, which drastically reduces SIRT efficacy. The objective of this project is to develop accurate patient-specific dosimetry for SIRT planning. We propose a novel method combining computational fluid dynamics (CFD) to simulate the 90Y microsphere 3D distribution and 90Y physics modeling to predict the absorbed dose. The central and novel approach is to carry out the CFD simulations for each patient's hepatic arterial tree to achieve high accuracy and precision, because anatomical features determining the microsphere distribution present wide variations across the patient population and prohibit the use of generic models. This novel CFD-based dosimetry will be the first comprehensive tool to integrate (1) the hepatic arterial tree extracted from the patient's standard-of-care angiogram, (2) CFD simulation in this hepatic arterial tree to predict and optimize the microsphere distribution, (3) calculation of the absorbed dose with 90Y physics modeling. Our long-term goal is developing a tool that can be integrated in clinical workflow to optimize the quantity and injection point of 90Y microspheres during SIRT planning. To this end, we will pursue two specific aims. (1) We will develop the CFD model and dose calculation using a pig model for validation; (2) we will develop a deep learning approach to simultaneously segment the hepatic artery from the standard-of-care patient angiograms and conduct a morphometric study of the obtained hepatic arterial trees to identify the principal parameters affecting the model. If successful, this project will generate a reliable, patient-specific dosimetry for SIRT providing a comprehensive calculation of the absorbed dose in individual lesions as well as in the healthy liver. This will enable high precision treatment planning to better treat the tumors with a ?dose-painting? approach and ultimately improve long-term patient outcome.